Novel image similarity metric for evaluating denoising and restoration techniques
Adrian Ciobanu, Tudor Barbu, Cristina Diana Niţă · 2017
A novel similarity metric is proposed here for evaluating the performance of various image denoising and restoration techniques. A testing methodology implying four types of noise contamination on medical images has been used to determine a vital threshold used in the computation of the proposed similarity index. The results are described along with values for known similarity indices. Graphical representations have been developed to allow a detailed investigation of the performance of different methods in specific image regions, like those in the proximity of the edges.